Status

We show the whole measurement.

You will not find polished traction on this page. batoon measures itself repeatably against a catalogue of 96 enterprise capabilities, eight layers and 13 industry scenarios. The result is here in full, including the third that is still missing.

As of 5 September 2026, after round 105. Previous measurement: 25 August 2026 at 51.2 %.

What runs

A substantial prototype, not a concept deck.

Three use cases end to end, across two tenants

A product launch for a B2B SaaS company. A churn rescue for an online retailer, with identity resolution, a learned churn model, per-contact decisions, a call centre under a capacity limit and vetoes from consent and suppression lists. A creative-ops loop from brief through three variants to approval. All three run in a curated simulated world.

Real agents, not a scripted demo

The runs orchestrate AI agents. There is also a deterministic mode with no model access. It makes the same demonstration reproducible without an API key.

Five real MarTech systems connected

Canva, Mailchimp, Meta Ads, HubSpot and Google BigQuery, each read-and-draft only. Nothing is sent, no campaign launches. A sixth configuration for WordPress.com is written but not yet verified against the real API.

Multi-tenant, with a real login

batoon is an OIDC relying party. It stores no passwords and runs no user directory of its own. Rights come from company and project role together: read, approve, configure. The browser round trip against a Keycloak broker has worked since 31 August 2026.

Operations: schedules, dialog, reportable matters

Runs start on request, on a schedule, through a reported event or at a scheduled time. Replies from contacts are decided without a run. Data access, erasure, consent withdrawal, legal threat and security incident go to a role with a deadline. All of it in the simulated world.

Scale measured in the simulated world

1,000,001 contacts pass through the whole churn-rescue chain in 180 seconds on one CPU. Every tool result stays between 5 and 700 bytes, at ten thousand contacts and at a million alike. The orchestrator sees numbers, never contacts.

The measurement

66.8 % of 96 capabilities.

66.8 %

625 of 936 weighted points · 13 of 13 industry scenarios feasible by our own catalogue · 0 critical must gaps · in August: 51.2 %, 2 of 13, 18 gaps

By layer

Decisioning 72.3 %
Orchestration & governance 70.8 %
Platform & operations 70.7 %
Design 69.0 %
Human–agent interaction 68.0 %
Distribution 65.0 %
Measurement & optimisation 62.5 %
Data layer 55.3 %

By criticality

Must, 49 capabilities, all at maturity 3 or higher 77.0 %
Should, 40 capabilities, 22 at maturity 3 or higher 51.2 %
Could, 7 capabilities, 1 at maturity 3 or higher 28.6 %

Maturity levels: 4 capabilities complete, 68 largely complete, 5 partial, 11 rudimentary, 8 without evidence. A scenario counts as feasible when all of its must capabilities reach at least maturity 3.

The 13 industry scenarios in the catalogue are not the three use cases above. The catalogue asks what a company in that industry would need overall. The use cases show what runs today. The jump from 51.2 % to 66.8 % comes from 46 rounds in ten days, measured in the simulated world. It says something about the mechanics. It says nothing about the effect on real customer data, because there is no live execution.

What is missing

The gaps, unvarnished.

Data layer, 55.3 %: Still the weakest layer. Identity resolution is deterministic, with no probabilistic matching and no golden record. On the live path the vendors keep their own IDs.
No live execution: The path is built and tested, but not armed. The five connected systems run read-and-draft only. The WordPress configuration is unverified against the real API.
No real customer data: Per-contact decisions, churn model, feedback loop, experiments and dialog run in the simulated world. What is measured there is mechanics. Impact on real revenue figures is still open.
24 capabilities below maturity 3: All in the should and could classes: 8 without evidence, 11 rudimentary, 5 partial. Among them the dialog without a language model, calls without a return channel, and schedules that cannot be edited in the interface.
No enterprise IdP in use: The browser round trip with PKCE runs against a Keycloak broker with a federated customer directory. It has never run against the identity provider of a real corporation.
One model provider evidenced: Six providers can be configured. Documented runs exist on Anthropic only. The fallback chain is tested, and its third stage depends on a valid key.
We set the assessment ourselves: The catalogue of 96 capabilities is ours. That makes the number internally comparable and reliable over time. The percentages per scenario are partly estimated from point differences, and the report says so itself. It does not replace external review.

Guardrail

What this page does not claim.

No live execution. Nothing is sent, no campaign launches.
No paying customers, no references, no case studies.
No results from production use. The use cases run in a simulated world.
No external review of the coverage figure. 13 of 13 means: by our own catalogue.

We name these limits because they would surface in any review anyway. The prototype does without such claims.